Some Discourse instances are setup as help resources and moderators are the ones often answering many questions as humans, despite search and FAQ facilities. Could there be a way for an ai persona to be styled on the basis of such a moderator’s historical interactions? So train the model on one user’s tone of voice, vocabulary and other syntactical aspects?
This could greatly assist the moderator’s job as the ai persona would be replying in a style very familiar to a longstanding existing user base.